From Policy to Practice: Progress on the Verifiable AI Governance Handbook
PreviewAI GovernanceEU AI ActManualOctober 15, 2026Reading time: 4 min.
Almost all companies that use AI already have a policy governing its use. Very few, however, could currently provide evidence that they are complying with that policy.
For two years, the conversation has centered on principles: transparency, human oversight, and risk management. Now the tone is shifting. The EU AI Act is already being implemented in phases; large clients are beginning to ask in their procurement questionnaires how their suppliers’ AI is controlled; and auditors are no longer satisfied with a signed document. The question has shifted from “Do you have a policy?” to “Prove it.”
On Tuesday, October 20, we published the Verifiable AI Governance Handbook, a 22-page practical guide on how to move from policy to proof. This article provides a preview of its contents and explains why we wrote it.
Three questions that will be brought before your committee
It doesn't matter what industry or what size the company is. When AI is involved in important decisions, sooner or later someone in a position of authority asks these three questions:
01
Who decides?
Who reviewed the AI's output, what authority did they have to change it, and how many times did they actually change it?
02
What does AI do?
What systems are in use, in which processes, what is the level of risk, and what data leaves the company's network perimeter?
03
What evidence is there?
Is there a record of all of the above, and can a third party verify it without having to take our word for it?
Most organizations can answer the first question with a name and the second with a spreadsheet. It's the third one they struggle with.
To state is not to prove
Stated Governance
It says what should happen
Policies, committees, risk matrices, and meeting minutes. They are necessary, but they merely describe an intention.
"Every AI-generated result in the hiring process is reviewed by a person."
Verifiable Governance
See what happened
Every revision leaves a record; every document is stamped, and anyone can verify it later.
"This report was reviewed on October 15 by someone with the authority to change it; they changed 2 out of 7 decisions; here is the evidence."
Governance isn't something you claim; it's something you demonstrate.
A Look Inside
Chapter 03 · Measurable Human Supervision
The manual suggests measuring supervision using a simple metric: the AI-to-human ratio. Of everything that AI suggests and a person reviews, what portion does the person change?
Example · Screening of Applications
71% AI29% human
The AI proposes a shortlist of 6 candidates. The reviewer discards one and adds another: 2 changes out of 7 items reviewed. A consistently low 0% over time is not good news: it usually indicates that the review is just a click away.
Chapter 07 · Verifiable Evidence
The manual includes a test that readers can try for themselves: two documents that are identical except for one number. One is sealed with V-Seal™; the other is not.
Original document
His fingerprint matches the one recorded when he signed it.
Certified documentCopy with a changed character
Just one digit is enough to make the fingerprint different.
Not found in the registryAnd to lead by example, the manual itself is watermarked: anyone who downloads it can verify that their copy is the original on the public verification portal—without needing an account and without having to upload the file.
What's Included
- 01Take Stockof Your AI
- 02ClassifyRisk Under the EU AI Act
- 03Human supervision: AI-to-human ratio
- 04Agentsand Chat: Pre-Submission Review
- 05Data: What Leaves Its Perimeter
- 06Codeand Models
- 07Verifiable Evidence
- 08ISO/IEC42001 and Controlled Documentation
- 09 90-Day Plan
- 10Appendices: Obligations, Templates, and Purchasing Checklist
Who is it for?
Guidance: What to check each quarter to ensure that AI is under control.
Technology: Chat controls, agents, code, and models that leave a paper trail.
Data and Privacy: What Leaves Your Network and What to Ask Each Provider.
Legal and Compliance: Each obligation under the EU AI Act is linked to specific evidence.
Launch
Tuesday, October 20.Free download.
22 pages, in Spanish, sealed with V-Seal™. We will publish it on this website and on the V-PROOF LinkedIn channel.
V-PROOF
The next generation of enterprise infrastructure is not AI.
It is Trust.
